[OpenIndiana-discuss] iostat - sum of the parts

[jason matthews]

I am aware of how this calculated for devices, but if underlying physical devices don't have I/Os why does the pool report I/Os taking so long?

In other words, the pool in this example reports and average wait time of 5170 ms but the devices report 0.0 save one which 0.1ms. Where does the delta come from?

j.

On Feb 10, 2014, at 6:36 PM, Joshua M. Clulow <josh at sysmgr.org> wrote:

> On 10 February 2014 18:02, jason matthews <jason at broken.net> wrote:
>> my question is why do the numbers for the zpool not resemble the sum of the parts? for example busy reports 56% for the pool but no device in the pool is more than 40% busy, in this example.
> 
> In both cases, the "% busy" column can be somewhat misleading.
> Essentially, this column represents the amount of time that we were
> waiting on the device to complete some number of I/O operations.  That
> is: for that percentage of the last refresh interval, we were waiting
> on _at least one_ I/O operation.  For the remainder of the interval
> the device was idle.
> 
> This applies to the "data" pool, as well: some amount of the time it
> is servicing I/O requests (the %b figure), and some amount of the time
> it is idle.  Critically, for "data" to be busy only _one_ of its
> underlying disks must be busy; busyness for "data" is the percentage
> of the last interval where _at least one disk_ had _at least one
> inflight I/O operation_.
> 
> -- 
> Joshua M. Clulow
> UNIX Admin/Developer
> http://blog.sysmgr.org
> 
> _______________________________________________
> OpenIndiana-discuss mailing list
> OpenIndiana-discuss at openindiana.org
> http://openindiana.org/mailman/listinfo/openindiana-discuss
>